sP7130133

 

Workshop on e-Science and High Performance Computing

eHPC2018

..

About eHPC

eHPC is a one-day seminar aiming to bring together a community of researchers, developers, practitioners, and user involved with e-Science and high performance computing technology from both academic institutions and industries to discuss and exchange ideas. The objectives of the workshop are to build collaborations for interdisciplinary research and to increase the awareness of technological advancement in e-science and HPC fields. The event will offer technical paper presentations, invited talks, discussions, and technology demonstrations by the e-Science and HPC communities.

Areas of interest include but not limited to e-Science applications, infrastructure and services; cloud computing; data analytic; scientific computing applications and tools. eHPC2017 is organized in conjunction with JCSSE 2018

..

Who can attend

Anyone who is interested in e-Science and High Performance Computing is welcome to join this seminar

Seminar Date and Venue

July 12, 2018, at 11:00-17:00 hrs.

Room IT321, Fl. 3, Faculty of ICT, Mahidol University, Salaya Campus, Nakhon Pathom, Thailand

(How to get here click)

..

Program for eHPC208

Time

Activity

8.30-10.40

JCSSE Keynote Speaker 1-2

1: Sentiment Analysis and its Application: Real-time Artificial

ProfHui-Huang Hsu, Tamkang University

2: Reconstruction and Evaluation of Software Architectures

ProfHorst Lichter, RWTH Aachen University

10.40-11.00

Coffee Break

11.00-11.30

eHPC opening ceremony

11.30-12.15

Keynote speaker: Dr. Sissades Tongsima

National Center for Genetic Engineering and Biotechnology, NSTDA

TopicThailand Genomics Big Data: the new kid on the block

12.15-13.00

Lunch

13.00-13.30

Assoc. Prof. Dr. Chantana Chantrapornchai and Asst. Prof. Dr. Putchong Uthayopas

Kasetsart University

Experience Sharing in International Student HPC Competition

13.30-14.00

Dr. Ekapol Chuangsuwanich

Chulalongkorn University

Topic: Deep learning and hardware: matching the demands from the machine learning community

14.00-14.30

Dr. Kanoksri Sarinnapakorn

Hydro and Agro Informatics Institute

Topic: Data Science driven weather forecasting solution

14.30-14.45

Coffee Break

14.45-15.15

Asst. Prof. Dr. Wannarat Suntiamorntut

Prince of Songkla University

Topic: Smart Connected Communities in Smart City: Phuket City (Case Study)

15.15-15.45

Dr. Wutthikrai Busayaporn

Synchrotron Light Research Institute

Topic: Surface Structure Determination: when Computation meet Experiment @ SLRI

15.45-16.15

Dr. Kanokrat Tiyapun

Thailand Institute of Nuclear Technology

Topic: High Performance Parallel Monte Carlo Transport Computations for Research Reactor Neutronics Applications

16.15-16.45

Dr. Suttipong Thajchayapong

National Electronics and Computer Technology Center, NSTDA

Topic: Toward Precision Poverty Alleviation using Thai Poverty Map and Analytics Platform (TPMAP) <<Download Slide>> 

16.45-17.00

Closing ceremony


Abstract

Thailand Genomics Big Data: the new kid on the block

Dr. Sissades Tongsima

National Center for Genetic Engineering and Biotechnology

National Science and Technology Development Agency

Dr.Sissades CV 

-------------------------------------------------------------

..

Experience Sharing in International Student HPC Competition

Assoc.Prof.Dr. Chantana Chantrapornchai and Asst.Prof.Dr. Putchong Uthayopas

HPCNC, Department of Computer Engineering,

Faculty of Engineering, Kasetsart University

อีเมลนี้จะถูกป้องกันจากสแปมบอท แต่คุณต้องเปิดการใช้งานจาวาสคริปก่อน , อีเมลนี้จะถูกป้องกันจากสแปมบอท แต่คุณต้องเปิดการใช้งานจาวาสคริปก่อน

..

Recent advancement in HPC technology creates a challenge in optimizing the next generation cluster of heterogeneous servers. These systems consist of many GPU servers linked together with a very high-speed interconnection network such as 10 Gb Ethernet or Infiniband. Moreover, the need to optimize the energy usage for a petascale system becomes substantially important. To inspire and stimulate the interest in optimizing such platform for an effective energy consumption, many world-class student cluster competitions have been created. These competitions such as SC student cluster competition, ISC student cluster competition and ASC is a great venue to enable computer engineering and computer science students to learn and practice about a very deep challenging HPC technologies. These include the tuning of hardware system, network, operating system, GPU, and the HPC application. In this presentation, the students experience in participating in two competitions; namely ASC2018 and ISC2018 will be shared. The challenges in the preparation and participation will be presented. It has been found that lack of a proper HPC test-bed system, continuous students training of various HPC systems and applications tuning techniques, and strong knowledge in parallel programming are the key obstacles to the graceful accomplishment. Thus,  having a  real HPC infrastructure in the university will benefit  students not only to gain  strong experiences in practicing for these aspects but it can serve  the HPC  research university community.

-------------------------------------------------------------

--

Deep learning and hardware: matching the demands from the machine learning community

Dr. Ekapol Chuangsuwanich

Department of Computer Engineering, Faculty of Computer engineering,

Chulalongkorn University

..

Deep learning or artificial neural networks have been gaining a lot of attention in the recent years, achieving state-of-the-art performance in many standard tasks such as automatic speech recognition, object recognition, and natural language processing. The formula for deep learning's success includes three key factors 1) the amount of data, 2) the computational capacity to handle the data, and 3) a large enough model capacity to handle the idiosyncrasy of the data. Larger and larger models with ever increasing amount of data have been developed. This significantly increases the amount of computation required for deep learning approaches. For example, AlphaGo Zero requires an estimate of 10^23 flops of computation to train. These kinds of compute requirement demands careful orchestration between software and hardware. In this talk, we will go over the techniques often employed in order to scaling up deep learning research. Additionally, we will also talk about an unconventional approach of using low-precision Logarithmic Number System (LNS) arithmetic for deep learning instead of the traditional float-based arithmetic which can greatly lower the hardware size and the power consumption per flop. With some modifications to the original back propagation algorithm, we were able to maintain the accuracy of traditional hardware with the LNS arithmetic. 

-------------------------------------------------------------


Data Science Driven Weather Forecasting Solutions

Dr. Kanoksri Sarinnapakorn

Hydro and Agro Informatics Institute

 

Weather has considerable impacts on ecosystems and lives of both humans and animals.  Knowing weather conditions can save lives and money.  Individuals use weather forecasts to plan day-to-day activities.  Businesses and organizations base their decisions on weather forecasts at some time or another.  Because of this, it is important to know the accurate situation of the weather so that informed decisions can be made.  Weather is an incredibly complex system and it illustrates today’s big data phenomenon. Weather forecasting has always been extremely challenging since the development of numerical weather prediction system (NWP) in 1950 due to the number of many variables involved and the complicated interactions between variables.  All numerical weather forecasts possess uncertainty, biases, and errors from inaccuracies in the specification of models describing physical processes, the choice of initial conditions, and the approximations used in solving the equations. Thus, NWP should not be the end of weather forecasting solution.  This talk presents some examples of how data science can generate more accurate weather forecasts.

Data science is an interdisciplinary field that blends knowledge in statistics and machine learning, computer skills, and domain expertise to extract meaningful insights and valuable information from data in various forms, structured and unstructured.  It uses a broad combination of scientific methods and algorithms to explore patterns and uncover trends in historical data, and then build predictive models to predict what might happen in the future.  Advancements in computing technologies together with the growing availability of weather-related data have dramatically improved the accuracy of forecasts. Researchers now invest more in High Performance Computing (HPC) systems in an effort to quickly uncover weather patterns in a vast volume of data.  Advanced statistical methods and machine learning algorithms such as clustering and classification methods are more robust to perturbations than physical models and they have been successfully applied to weather prediction. Decision tree, deep learning, reinforcement learning, artificial intelligence, K-NN (K-nearest neighbors) are some other methods used in studying weather models and forecasting some weather events such as tracking movement, speed and ferocity of storms, rainfall prediction, predicting monsoon, radar rainfall bias correction, etc.  Data assimilation exploits knowledge of weather prediction model and observation uncertainties to optimize model prediction.  An increase in accuracy of weather forecasts is achieved through an iterative process that uses recent/current observations to adjust a model forecast. Therefore, while we are not able to control the weather, an effective use of data science can help us discover new knowledge in the weather system and provide a more reliable solution to weather forecasting.  

-------------------------------------------------------------

.

Smart Connected Communities in Smart City: Phuket City (Case Study)

W.Suntiamorntut, S.Charoenpanyasak, J.Nopparat, M.Manopirulporn and S.Julrat

Computer Engineering Department, Prince of Songkla University, Hatyai Songkhla 90112 Thailand

อีเมลนี้จะถูกป้องกันจากสแปมบอท แต่คุณต้องเปิดการใช้งานจาวาสคริปก่อน

.

This paper introduces the first smart city model; Phuket Smart city, according to Thai Government to drive Digital Economy in Thailand. Prince of Songkla University together with Digital Economy Promotion Agency (DEPA), Ministry of Digital Economy and Society, CAT Telecommunications Company and National Electronics and Computer Technology Center, Ministry of Science and Technology were collaborated and established Phuket Smart City Project in 2017. As we have known, Phuket is the world class tourism, more than 13 million people visited us every year. In order have a good management system for Phuket province, the smart systems have been proposed under Phuket smart city project. In addition, we also aim to add-up high value tourism in the city. Thus the important issue is to improve the infrastructure of connectivity which allows the things (mobile phone, sensors, etc.) in the city connected. When we have enough information or data gathering from things around the city, we are going to provide a high standard service related to the behavior of our tourism. Therefore, we set a development architecture as shown in Figure1. The first three topics we focused are Economic, People and City Management. The main economic of Phuket city is tourism and we expect to be a global digital hub. In order to draw many professional careers around the world to live in Phuket, we have to concern about their wealthy, healthy and safety. In addition, we have to take care our beautiful environment using smart IoT system.

Prince of Songkla University (Computer Engineering Department) has responsibility to install our technology in five smart systems consists of City Data Platform, Tourist mobile application, Marine safety, iLertYou and IoT Environment sensors. All data from those system, connected to the city data platform as shown in Figure 2. We have enabled our in-house IoT devices such as LoRa, NB-IoT (AIS), Zigbee and WiFi via PSU-IoT Things Platform. Moreover, all information from mobile application and CCTV are streaming to our city data platform. The analytic services are also implemented here in this layer. Finally, cross analytic will be applied and created a new business model in the city. 

-------------------------------------------------------------

.

Surface Structure Determination: when Computation meet Experiment @ SLRI

1*Dr.Wutthikrai Busayaporn, 2Dr.Ittipon Fongkaew, 3Dr.Wutthigrai Sailiam, and 1Dr.Maneerat Chotsawat

 .

1Synchrotron Light Research Institute (Public Organization), Nakhon Ratchasima

2School of Physics, Suranaree University of Technology, Nakhon Ratchasima

3Department of Applied Physics, Faculty of Engineering, Rajamangala University of Technology ISAN (Khon Kaen Campus), Khon Kaen

* อีเมลนี้จะถูกป้องกันจากสแปมบอท แต่คุณต้องเปิดการใช้งานจาวาสคริปก่อน

.

Surface sciences abrupt the knowledge on solid state physics and materials sciences rapidly even the fundamental the of surface science itself based on pieces of information from both topics above. Reasons behind these are behavior of surface interaction, structure and mechanism most likely be different from the same materials when it is in bulk. However, the drawback of investigation of surface sciences are the complications of instruments and theory. Synchrotron Light Research Institute, SLRI, situated with synchrotron radiation facility and attached beamlines which few of them dedicated for surface science experiments. Various techniques such as X-ray Absorption Spectroscopy (XAS), Photoemission Electron Spectroscopy (PES) and Photoemission Electron Microscopy (PEEM), can play important roles to elucidate big data from surface structure. Nevertheless, such interactions and mechanisms can still be questioned. High Performance Computing (SLRI-HPC) established to be an extremely important jigsaw to link between experiment, theory and real applications. By Computational Materials Physics (CMP) project using tools from Density Functional Theory (DFT) to Molecular Dynamics (MD). CMP project become the support unit for SR-experiments in many cases such as calculated the band structure and band gap of oxide semiconductors, discover the behavior of thin film growth, reveal the mechanism of adatoms on surface and understand efficiency of catalyst. Moreover, SLRI-HPC soon will extend its capabilities to larger disciplinary such as design of magnetic lattice, 3D reconstruction of tomography and even drug design. 

-------------------------------------------------------------

.

High Performance Parallel Monte Carlo Transport Computations for Research Reactor Neutronics Applications

Dr. Kanokrat Tiyapun

Thailand Institute of Nuclear Technology

..

High performance computing (HPC) is used for various problems required intense computation. It can be applied to large scale neutronics calculations for nuclear safety which is required to support design activities for the fuel management and burnup calculation of the nuclear research reactor. The complexity of reactor geometry configurations, control system, fuel burnup and fission products are essential challenges for neutronics calculations. Monte Carlo codes for particle transport can take advantage of accessible parallel computing because of the inherently parallel nature of the computational algorithm. The most common approach to achieve high performance particle transport Monte Carlo is to utilize particle parallelism using a master/slave algorithm. With this algorithm, the master process will distribute individual histories to different slave computing nodes. The Message Passing Interface (MPI) is used to distribute problem data to the slaves and collect the tallied results, while Open Multi-Processing (OpenMP) is used for threading computation within each slave. The neutron tracks can be processed from a source throughout its life in parallel without heavy load on communication between slave nodes due to the independence of neutron particle histories. Using parallel computations on high performance cluster computers significantly increase number of sampled particles and therefore produce small variance, high reliable confidence interval and reduce the relative error in the computed statistic. The Monte Carlo N-Particle Transport Code (MCNP5) parallel performance which is assessed on the HPC has been proven to be effective approach for nuclear research reactor problems. 

-------------------------------------------------------------

...

Toward Precision Poverty Alleviation using Thai Poverty Map and Analytics Platform (TPMAP)

Dr. Suttipong Thajchayapong

Computational Process Analytics (CPA) Lab

Data Analytics and Computing Research Unit, NECTEC

..

In this talk, we present Thai Poverty Map and Analytics Platform or TPMAP, a project initiated by the Thai government with the ultimate goal of precision poverty alleviation. Developed by the Office of National Economic and Social Development Board (NESDB) and National Electronics and Computer Technology Center (NECTEC), TPMAP enables policy-makers to identify the poor in their respective areas, locate them and understand their basic needs. TPMAP integrates data sources from two government agencies: 1) a census-based BMN data source of approximately 36 millions individuals from the Community Development Department, Ministry of Interior and 2) a register-based data source of approximately 14 millions individuals from the Ministry of Finance. First, TPMAP uses Multidimensional Poverty Index (MPI) to identify the MPI-poor individuals from the BMN data source based on their lack of five basic needs (education, health, living standard, income and access to public services). Then, this MPI-poor group is intersected with the second registered-based data source where a target group of 1.4 millions are identified. It is shown that these target individuals in different areas indeed have different needs and it is possible to formulate precision poverty alleviation programs.

-------------------------------------------------------------